The ISME Journal (2015) 9, 1246–1259 & 2015 International Society for Microbial Ecology All rights reserved 1751-7362/15 www.nature.com/ismej ORIGINAL ARTICLE Culture and molecular-based profiles show shifts in bacterial communities of the upper respiratory tract that occur with age

This article has been corrected since Advance Online Publication and an erratum is also printed in this issue

Jennifer C Stearns1, Carla J Davidson2, Suzanne McKeon2, Fiona J Whelan3, Michelle E Fontes1, Anthony B Schryvers2, Dawn ME Bowdish4, James D Kellner2,5 and Michael G Surette1,2,3 1Department of Medicine, McMaster University, Hamilton, Ontario, Canada; 2Department of Microbiology, Immunology and Infectious Diseases, University of Calgary, Calgary, Alberta, Canada; 3Department of Biochemistry and Biomedical Sciences, McMaster University, Hamilton, Ontario, Canada; 4Department of Pathology and Molecular Medicine, McMaster University, Hamilton, Ontario, Canada and 5Department of Paediatrics, University of Calgary, Calgary, Alberta, Canada

The upper respiratory tract (URT) is a crucial site for host defense, as it is home to bacterial communities that both modulate host immune defense and serve as a reservoir of potential pathogens. Young children are at high risk of respiratory illness, yet the composition of their URT microbiota is not well understood. Microbial profiling of the respiratory tract has traditionally focused on culturing common respiratory pathogens, whereas recent culture-independent microbiome profiling can only report the relative abundance of bacterial populations. In the current study, we used both molecular profiling of the bacterial 16S rRNA gene and laboratory culture to examine the bacterial diversity from the oropharynx and nasopharynx of 51 healthy children with a median age of 1.1 years (range 1–4.5 years) along with 19 accompanying parents. The resulting profiles suggest that in young children the nasopharyngeal microbiota, much like the gastro- intestinal tract microbiome, changes from an immature state, where it is colonized by a few dominant taxa, to a more diverse state as it matures to resemble the adult microbiota. Importantly, this difference in bacterial diversity between adults and children accompanies a change in bacterial load of three orders of magnitude. This indicates that the bacterial communities in the nasopharynx of young children have a fundamentally different structure from those in adults and suggests that maturation of this community occurs sometime during the first few years of life, a period that includes ages at which children are at the highest risk for respiratory disease. The ISME Journal (2015) 9, 1246–1259; doi:10.1038/ismej.2014.250; published online 9 January 2015

Introduction In addition to its beneficial role, the commensal microbiota can also be a reservoir of respiratory As the point of entry to both the lower respiratory pathogens as well as antibiotic resistance and tract and the gastrointestinal tract, the upper virulence genes (Garcia-Rodriguez and Fresnadillo respiratory tract (URT) is continuously exposed to Martinez, 2002; Liu et al., 2013). For instance, the outside world. The microbiota is involved in healthy people can have long periods of asympto- resistance to colonization by incoming pathogens matic carriage of classic respiratory pathogens (Kelley et al., 2005; Margolis et al., 2010), education (Garcia-Rodriguez and Fresnadillo Martinez, 2002) of the immune system (Lathrop et al., 2011) and and invasive disease can often arise from normally regulation of host immunocompetence in the lung benign bacterial residents (Bakaletz, 2004). This in response to infection (Ichinohe et al., 2011). duality blurs the definition of a commensal member of the respiratory tract. Instead, it’s likely that this body site harbours an indigenous microbiota whose Correspondence: MG Surette, Department of Medicine, McMaster members behave differently, depending on factors University, 1200 Main St West, Hamilton, Ontario L8S 4L8, such as their location in the body (Blaser and Canada. E-mail: [email protected] Falkow, 2009), bacterial community disturbance Received 11 July 2014; revised 18 November 2014; accepted (Lynch, 2013), environmental pressures (Feldman 24 November 2014; published online 9 January 2015 and Anderson, 2013) and/or immune responses in Changes in the microbiota of the upper airways with age JC Stearns et al 1247 the host (Starkey et al., 2013). Unlike many acute order to describe the healthy child URT microbiome infectious diseases where a single microbe can be in the context of the adults with whom they have the targeted and eradicated, lung infections are often most contact. We found that the nasopharynx of polymicrobial (Bakaletz, 2004; Han et al., 2012; young children is dominated by a small number of Huang et al., 2012, 2014; Dickson et al., 2013) and bacterial groups that are present in high total the organisms recovered from respiratory and numbers, in contrast to that in adults, which had invasive infections are often a mixture of common much lower bacterial carriage and a more diverse URT microbes (Laupland et al., 2000, Sibley et al., community. The oropharynx communities 2008; 2012). Respiratory infections have a higher were dominated by streptococci in all subjects and impact on health worldwide than all other infec- with higher bacterial biomass than the nasopharynx. tious diseases combined (Mizgerd, 2006) and mor- This study provides the first comprehensive look at tality rates associated with lung infections have not the healthy URT microbiota of young children along significantly improved in over 50 years (Mizgerd, with those of their parents. 2008). Children, in particular, are very susceptible to respiratory illness (Liu et al., 2012), therefore, an Materials and methods understanding of the makeup of the URT microbiota in this population is important to our understanding Cohort of the origin and progression of infection. Interac- The Calgary Area Streptococcus pneumoniae Epi- tions between resident airway pathogens are known demiology Research (team has conducted 10-point to occur, for instance, competitive exclusion prevalence surveys of pneumococcal NP coloniza- between Streptococcus pneumoniae and Staphylo- tion in healthy children o5 years of age attending coccus aureus (Lijek and Weiser, 2012); however, Community Health Centres for routine immuniza- these interactions have not been studied in the tion visits in Calgary, Canada since 2003 (Kellner context of the complete URT microbiota. et al., 2008; Ricketson et al., 2014). The study was Until recently, the focus of most URT microbiol- approved by the University of Calgary Conjoint ogy in healthy individuals has been on carriage of a Health Research Ethics Board. Each survey was few important pathogens such as S. pneumoniae, conducted over approximately a 6-week period in Haemophilus influenzae, Moraxella catarrhalis and seven Community Health Centres. After obtaining S. aureus (Dunne et al., 2013), however, our knowl- written informed consent, study nurses obtained edge of colonization and succession of bacterial demographic information and conducted a health communities in the healthy human URT is lacking. survey. An average of 615 children were enrolled in A few microbiome studies have focused on describ- each survey. During the 2011 and 2012 surveys, ing the communities at these sites (Lemon et al., additional written informed consent was obtained 2010; Charlson et al., 2011; Faust et al., 2012), for a convenience subset of the study population to however, only Bogaert et al. (2011) looked in depth provide additional samples that were used in the at children under 2 years of age, the population at present study to describe the healthy child URT greatest risk of respiratory illness and invasive microbiome. NP swabs were taken nasally and OP pneumococcal disease. No reports, however, have swabs orally according to the standard WHO looked at the healthy URT microbiota from both the method (O’Brien and Nohynek, 2003) by the public perspective of molecular profiling and quantitative health nurse from 51 children and 19 adults culture for all bacteria, an approach that provides a (Supplementary Table S4). We used Copan eSwabs more complete picture of bacterial colonization. (Alere Canada, Ottawa, ON, Canada), a flocked swab Molecular profiling can provide a snapshot of with 1 ml of Amies transfer fluid. All swabs were community structure, however, this picture is stored at room temperature and processed within imperfect because of its inability to distinguish 6–8 h. For processing, the swabs in their transfer between living cells and dead cells or cell-free fluid were vortexed vigorously for 15–30 s, then a DNA. It also suffers from bias against some bacterial 100 ml subsample was taken for plating. The remain- lineages, as no primer pair is perfect, and a lack of der was frozen at À 20 1C for molecular analysis. taxonomic resolution, as relatively short DNA sequences are used. Bacterial DNA isolation and Illumina sequencing of For the study herein, a subset of the swabs bacterial tags collected during a large point prevalence study of DNA was extracted from NP and OP swab samples S. pneumoniae serotype carriage in the Calgary area with a custom DNA extraction protocol involving (Ricketson et al., 2014) were used. A total of 51 mechanical and enzymatic lysis followed by a phenol:- healthy children, with a median age of 1.1 years, chloroform extraction and a clean-up step. First, 300 ml from those attending community health centers for of sample was added to a tube containing 0.1-mm glass routine immunization, along with 19 accompanying beads (MoBio Laboratories Inc., Carlsbad, CA, USA) parents had nasopharyngeal (NP) and oropharyngeal along with 800 mlof200mM sodium phosphate (OP) swabs collected. Molecular and quantitative monobasic (pH 8) and 100 ml guanidinium thiocyanate culture profiles of the bacterial communities within EDTA N-lauroylsarkosine buffer (50.8 mM guanidine the oropharynx and nasopharynx is presented in thiocyanate, 100 mM ethylenediaminetetraacetic acid

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1248 and 34 mM N-lauroylsarcosine). These were then (Ye, 2011), with a clustering threshold of 97%. homogenized in the PowerLyzer 24 Bench Top Output from this tool was then formatted for input Homogenizer (MoBio Laboratories Inc.) for 3 min at into QIIME, where was assigned using the 3000 revolutions per minute. Next, two enzymatic Ribosomal Database Project classifier (Wang et al., lysis steps were performed. In the first, the sample 2007), with a minimum confidence cutoff of 0.8 was incubated with 50 ml of 100 mg ml À 1 lysozyme, (QIIME default) against the Greengenes (4 February 500 U mutanolysin and 10 mlof10mgmlÀ 1 RNase for 2011 release) reference database to the genus level 1 hour at 37 1C. In the second, the sample was (DeSantis et al., 2006). All OTUs classified as incubated with 25 ml 25% sodium dodecyl sulphate, ‘Root:Other’ were excluded (Supplementary Table 25 mlof20mgmlÀ 1 Proteinase K and 62.5 mlof5M S1) along with three samples with poor sequence NaCl at 65 1C for 1 h. Next, debris was pelleted in a coverage and aberrant microbial profiles. tabletop centrifuge at maximum speed for 5 min and The a-diversity measures used were observed the supernatant added to 900 ml of phenol:chloro- OTUs, which is an estimate of community richness, form:isoamyl alcohol (25:24:1). The sample was and Shannon diversity, which is an estimate of then vortexed and centrifuged at maximum speed community diversity based on richness and evenness in a tabletop centrifuge for 10 min. The aqueous of the community (Shannon and Weaver, 1949). phase was removed and the sample run through the b-diversity, which is a measure of the differences Clean and Concentrator-25 column (Zymo Research, between communities, was calculated with the Irvine, CA, USA) according to kit directions except weighted UniFrac distance and clustering of the for elution, which was done with 50 ml of ultrapure samples based on distance was illustrated with water and allowed to sit for 5 min before elution. principal coordinate analysis. OTUs containing a The DNA was quantified using a Nanodrop 2000c single sequence (singletons) were removed prior to Spectrophotometer. b-diversity calculations. Weighted UniFrac distance Amplification of bacterial 16S rRNA gene v3 was calculated on evenly sampled profiles (sample region tags was done as in (Bartram et al., 2011) depth of 2400 sequences, without replacement). The with the following changes: 5 pmol of primer, 200 mM number of clusters were evaluated with the gap À 1 of each dNTP, 1.5 m M MgCl2,2mlof10mgml statistic, which provides an estimation of how well bovine serum albumin and 1.25 U Taq polymerase each k, or number of clusters, fits the data. Gap (Life Technologies, Carlsbad, CA, USA) were used in statistic estimation was done with 500 Monte Carlo a50ml reaction volume. The PCR program used was samples in R with the use of the phyloseq (McMurdie as follows: 94 1C for 2 min followed by 30 cycles of and Holmes, 2013) clusGap wrapper. In order to 94 1C for 30 s, 50 1C for 30 s and 72 1C for 30 s, then a resolve sample-clustering jackknife analysis using final extension step at 72 1C for 10 min. Illumina weighted UniFrac distance was performed on evenly libraries were sequenced in the McMaster DNA sampled OTU tables (10 tables, 2400 sequences per Sequencing Facility with the following steps. sample) and a composite unweighted pair group Pooled libraries were first tested on an Agilent method with arithmetic mean (UPGMA) tree of the BioAnalyzer High Sensitivity DNA chip and then samples was created. a-andb-diversity calculations quantified with qPCR using Illumina’s PhiX control and plots were done in R with the phyloseq package, library as a standard, SYBR fast 2 Â qPCR mastermix except for jackknifed UPGMA trees, which were done (KAPABiosystems, Wilmington, WA, USA) and pri- in QIIME version 1.8.0 (Caporaso et al., 2010). mers that bind to the distal ends of the adaptors Taxonomic summaries and multiple response per- (flowcell-binding regions): P5 50-AATGATACGGC mutation procedure with 1000 permutations, were GACCACCGA-30,P750-CAAGCAGAAGACGGCATA also both done with QIIME. A w2-test of each age CGA-30. 16S rRNA gene v3 region pools were then group among NP clades (Ib, IIa and IIb) was done in R combined with PhiX control DNA in a 9:1 ratio and with 500 P-value simulations. OTUs that differed 250 bp were sequenced in the forward and reverse significantly between adult and child samples were direction on the Illumina MiSeq instrument. The selected from NP and OP data separately using LefSe, completed run was demultiplexed with Illumina’s which combines a non-parametric Kruskal–Wallis Casava software (version 1.8.2). sum-rank test to determine which OTUs have a different abundance between ages, a unpaired Wil- Sequence processing and data analysis coxon rank-sum test to determine which age is Custom, in-house Perl scripts were developed to statistically relevant and effect size is estimated using process Illumina sequences and are available from linear discriminant analysis (Segata et al., 2011), a the authors. First, Cutadapt (Martin, 2011) was used default cutoff of Po0.05 and linear discriminant to trim any reads that exceeded the v3 region of the analysis effect of 42.0 was used. OTUs with a total bacterial 16S rRNA gene. The resulting paired-end relative abundance across all samples of o0.11% sequences were aligned with PANDAseq (Masella were removed from the data prior to LefSe analysis, et al., 2012) and the sequences with any mismatches as were samples from children over 19 months age or ambiguous bases were culled. Input sequences (all of which were between 4 and 4.5 years of age) from all samples were clustered into operational owing to the small relative number of samples from taxonomic units (OTUs) using Abundant OTU þ this group (6/69 for NP and 4/49 for OP). All OTU

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1249 comparisons between groups were corrected for sequenced in the forward direction by Beckman multiple testing using the false discovery rate Coulter Genomics (Danvers, MA, USA). Low-qual- (Benjamini and Hochberg, 1995) with the p.adjust ity regions of each sequences were trimmed for 1% function in R. error probability in Geneious (v5.6.6) then taxon- omy was assigned by BLAST alignment (Altschul Streptococcus phylogeny et al., 1990) to the Human Oral Microbiome In order to resolve the species distribution of Database and counts were compiled into OTUs and cultured isolates assigned to the Supplementary Table S7. All e-values were below Streptococcus genus, a phylogenetic placement 7.26e–115 over the entire length of the partial 16S method was used, whereby these short sequences rRNA gene sequences used. were placed onto a reference tree made from reference strains of Streptococcus species from the Human Oral Microbiome Database (Chen et al., Results 2010). All full-length 16S rRNA gene reference sequences for Streptococcus species (29 in total) Bacterial communities within the nasopharynx and from Human Oral Microbiome Database were oropharynx are distinct from one another aligned using MUSCLE (multiple sequence com- The development of the URT microbiota is poorly parison by log-expectation; Edgar, 2004) then used understood, although some studies have described to create a maximum likelihood phylogeny with pathogen carriage in healthy people, few have RaXML (randomized axelerated maximum like- looked at the commensal microbiota in healthy lihood; Stamatakis, 2006; Silvestro and Michalak, young children. To investigate the microbiome of 2012) using the Generalized Time Reversible þ young children, we obtained NP and OP swabs from Gamma evolutionary model þ a proportion of (I) healthy children during routine immunization in invariable sites and the rapid bootstrap method Calgary, AB, Canada. Most of the children were over 10 000 reps (Supplementary Figure S1A). between 12 and 19 months, with a small subset aged A separate phylogeny created with MrBayes between 4 and 5 years. The parents of these children (Ronquist and Huelsenbeck, 2003) using the same were asked to volunteer samples, and in total, we evolutionary model (GTR þ G þ I) over 10 million analyzed NP swabs from 69 participants (51 from generations had an identical topology to the children and 18 from adults and OP swabs for 34 maximum likelihood tree with better confidence children and 15 adults, for a total of 118 samples, all for branching (Supplementary Figure S1B). Varia- except two of which were matching (Supplementary tion in likelihood values visualized with Tracer Table S4). v1.6 (Rambaut and Drummond, 2009) were used to Ecological diversity measures were used to dis- confirm convergence of the bayesian tree. Short tinguish differences in bacterial 16S rRNA gene sequences were placed onto the maximum like- molecular profiles from the nasopharynx and the lihood phylogeny with pplacer (Matsen et al., oropharynx in healthy children and adults. 2010). Confidence in each placement was mea- Figure 1a shows a-diversity estimates, which pro- sured with expected distance between placement vide an indication of how the bacterial communities locations (Supplementary Table S8) and was good in each sample were structured. The number of for most OTU sequences. observed OTUs is a straightforward count of the number of unique OTUs in each sample and Shannon diversity is an estimate of community Quantitative culture diversity based on the richness and evenness of all Quantitative culture for NP swabs from 20 children OTUs within a sample. Both the observed number of and five of their parents and OP swabs for four of OTUs and Shannon diversity estimates for NP these adult–child pairs was done using five media samples had a broader range than did those for OP types (BHI CO, brain heart infusion supplemented samples. Also, adults had a significantly higher with colistin sulfate (10 mgmlÀ 1) and oxolinic acid number of observed OTUs and higher Shannon (5 mgmlÀ 1); CBA, Columbia blood agar; Chocolate diversity in the nasopharynx than did the children agar; McKay agar; and MSA, mannitol salt agar) and o19 months of age at either site swabbed (Po0.05). incubated aerobically at 37 1C with 5% CO2.Three Together, this suggests that bacterial communities in dilutions (10 À 1,10À 3 and 10 À 5)wereplatedon the adult nasopharynx were more diverse than those each media type and incubated for 48 h. Morpho- in the nasopharynx or the oropharynx of young logically similar colonies from each plate were children. counted and used to calculate colony-forming unit b-diversity estimates complement a-diversity by (CFU) ml À 1 for each isolate. A representative illustrating the difference in community member- isolate from each type and media was picked and ship between samples and the weighted UniFrac stored in 10% skim milk at À 80 1C. The 16S rRNA distance shown in Figure 1b uses both the abun- gene from each of 4300 cultured isolates was dance and phylogenetic relationship of the OTUs in amplified using the 8F–926R primers (Liu et al., each sample to determine how similar samples are 1997), and the resulting 700–900 bp product was to one another. We suspected that the relative

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1250

Observed OTUs 0.1 300 * * 0.0 200 −0.1

100 [17.8%] Axis.2 −0.2 NP OP 0 −0.2 0.0 0.2 Shannon diversity Axis.1 [54%] 8 * Child(<19mo) * 0.2 Child(>4yrs) 6 Adult Alpha diversity measure 0.1 4

2 0.0

0 −0.1 Child Child Child Child [10.9%] Axis.3 Adult Adult (<19mo) (>4yrs) (<19mo) (>4yrs) Nasopharynx Oropharynx −0.2 −0.2 0.0 0.2 Axis.1 [54%] Figure 1 Oropharyngeal and nasopharyngeal communities are distinct. (a) Alpha diversity of oropharyngeal and nasopharyngeal microbial communities in adults and children have a range of values, however, adult NP samples appear to have higher diversity than OP or child NP samples. Whiskers are from the 10–90 percentile and statistical significance are for Po0.05 (Kruskal–Wallis and Dunn’s multiple comparison test). (b) Beta diversity of the microbial profiles within the oropharyngeal and nasopharyngeal samples show separation by sample site. The first three principal coordinates of a principal coordinate analysis, based on weighted UniFrac distance, shows distinct clustering of samples by swab geography as opposed to age of the subject. Tight clustering of adult and child OP samples suggests similarity in these profiles, whereas scattering of NP samples suggests dissimilarity in the profiles.

abundance of bacterial groups would be essential for (Figure 2a). These plots show the diversity differ- describing differences between samples and indeed ences calculated using a-diversity estimates jackknife b-diversity showed good reproducibility (Figure 1a). Note that bacterial groups within each for weighted measures, weighted UniFrac and Bray– profile were identified down to the best classifica- Curtis (not shown), and less confidence for tion possible on the basis of taxonomic information unweighted UniFrac distances (Supplementary contained within the v3 region of the 16S rRNA Figure S2). Of the metadata available, only swab genes, hence many bacterial groups are labeled with site and age corresponded to separation of samples the genus and others by only the family or order in the principal coordinate analysis (Figure 1b). name. The oropharynx of both children and adults Bacterial communities from the nasopharynx or not only had a high proportion of Streptococcus but oropharynx were distinctly separated from each also contained bacterial groups often cultured from other in the ordination in Figure 1b and were this site such as Rothia, Prevotella, Gemella, significantly different from each other based on a Veillonella, Fusobacteria, Haemophilus and Neis- multiple response permutation procedure (A ¼ 0.25, seria (Figure 2 and for a complete list see significance of delta ¼ 0.001), indicating that dis- Supplementary Table S2). NP samples from adults tinct bacterial communities likely colonize each site. had a large proportion of such as It is also apparent that OP bacterial communities in Lachnospiraceae, Staphylococcus and Streptococ- different people were all very similar to one another, cus, Bacteriodetes such as Sphingobacterium and compared with the heterogeneity in the NP samples, Prevotella and Actinobacteria other than Coryne- as distances were smaller between any two OP bacterium such as Bifidobacteria, Rothia and Pro- samples (0.127±0.054) than between any two NP pionibacterium. Child NP bacterial communities samples (0.276±0.095) (Po0.0001; Supplementary contained similar groups but had larger proportions Figure S3). of Proteobacteria such as Moraxella, Enterobacter- iaceae and Haemophilus, as well as the Firmicute Enterococcus (Figure 2 and for a complete list see Adults and children have different bacterial Supplementary Table S3). communities in the nasopharynx To determine whether there was distinct cluster- The relative taxonomic profiles for adult or child ing of samples, as suggested by Figure 1b, weighted samples from each site were averaged in order to UniFrac distances calculated above were used to clearly illustrate the differences between them determine how well estimates of each of 1–10

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1251 abMycoplasma Nasopharynx Oropharynx Pseudomonas Moraxella Haemophilus Raouletta Escherichia 0.8 Nisseria Leptotrichia Fusobacterium Moraxella Veillonella 0.6 Faecalibacterium Lachnospiraceae Anaerococcus Streptococcus 0.4 Lactobacillus Gap 0.2 Gemella Staphylococcus Enterococcus Sphingobacterium 0.0 Prevotella Bacteriodes Bifidobacterium 12345678910 Propionibacterium Porphyromonas k Rothia AdultChild Corynebacterium Adult Child Actinomyces c I II 1 Ia Ib IIa IIb 1 1 0.9 0.06 0.9 0.9 0.7 1 1 0.8 0.9 0.9 1 1 1 0.8 0.8 1 0.9 0.9 0.8 1 1 1 0.7 1 0.8 0.9 0.8 1 0.8 1 0.8 0.7 1 1 0.7 0.7 1 1 1 0.8 0.7 1 1 1 0.9 0.7 0.9 1 1 0.9 0.9 0.9 1 1 1 1 d

Child(<19mo) Child(>4yrs) Adult

Oropharynx Nasopharynx Figure 2 Taxonomic profiles illustrate the differences between groups. (a) Microbial community profiles, summarized down to the genus level when possible, for adults and children at both sample sites. (b) Estimation of the gap statistic for each of 1–10 clusters of samples, based on weighted UniFrac distances. (c) UPGMA tree of weighted UniFrac distances with jackknife support (only values above 0.6 are shown in the tree). (d) Taxonomic profiles were ordered based on their position in the UPGMA phenogram labeled by age and sample site (colours are the same as in a). Similar colour groups were used for each phyla: Actinobacteria (purple/pink), Bacteriodetes (orange), Cyanobacteria (grey), Firmicutes (green/yellow/brown), Fusobacteria (red), Proteobacteria (blue), and Tenericutes (black). A complete list of taxa is available in Supplementary Tables S2 (OP) and S3 (NP). Red circles in d indicate NP samples that were grouped with OP samples. clusters fit the data with a gap statistic calculation. Separation of samples into clades and subclades Gap statistic values increased steadily then level off was due to taxonomic differences within them, as for k ¼ 4, suggesting that there were four clusters of UPGMA was based on the weighted UniFrac samples (Figure 2b). Differences between profiles distance between samples. For instance, clade I were illustrated using UPGMA hierarchical cluster- contains profiles with a high proportion of Firmi- ing with jackknife support, also based on weighted cutes, whereas clade II contains samples with a high UniFrac distances. The resulting tree, shown in proportion of Gram-negative bacterial taxa as well as Figure 2c, branches into two main clades (I and II) Actinobacteria. Subclade Ia includes all bacterial and these each branch again into two subclades (Ia, profiles from the oropharynx, as well as three NP Ib, IIa and IIb), separation of which had good samples that likely grouped with them owing to the jackknife support. Taxonomic profiles for each high proportion of Streptococcus. Subclades Ib, IIa sample were ordered according to the UPGMA and IIb contain all of the NP swab samples, with the phenogram (Figure 2d) and the pattern that emerged exception of the previous three. Subclade Ib was the illustrates the differences between samples that most taxonomically diverse, whereas subclades IIa were suggested in Figure 1. was dominated by Actinobacteria (specifically

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1252 Corynebacterium) and/or Firmicutes (specifically Bacterial OTUs differ significantly between adult and Carnobacteriaceae) and subclade IIb was dominated child microbial communities by Proteobacteria (Moraxella). A small number of Bacterial OTUs, rather than taxa summaries, were samples (4/41) had all three taxonomic groups. A analyzed with LefSe to identify statistically signifi- subset of subclade IIb (3/22) had a substantial cant differences between adult and child microbial proportion of Haemophilus without Streptococcus profiles, as OTUs with the same taxonomic assign- and another subset (3/22) with the opposite compo- ment sometimes varied between adults and chil- sition. From this ordering of samples, we can see dren. Owing to the small number of samples from that subclades containing NP samples clustered by older children (4/49 for OP and 6/69 for NP), these age of the subjects (w2, P ¼ 0.002, over that expected were excluded from the analysis. In the oropharynx, by chance). Subclade Ib was enriched with samples microbial communities in young children had a from adult nasopharynx and subclades IIa and IIb significantly higher proportion of 91 different OTUs, were enriched with samples from the nasopharynx including Streptococcus (OTU3), Gemella (OTU 14), of young children (o19 months). Of note is the Haemophilus (OTU 15), Neisseria (OTU 6), Porphyr- proportion of the children over the age of 4 years omonas (OTU 30), Granulicatella (OTU 24 and 25), that were in subclade Ib (4/6). an unclassified Fusobacteriaceae (OTU 36), among To test whether adults and children from the others (Figure 4a). Interestingly, several different same family had URT microbial communities that Prevotella OTUs were significantly different were more similar to each other than to other between adults and children, with some more people, weighted UniFrac distances were calcu- prevalent in children and others in adults. Microbial lated between each sample and shown for each communities in the oropharynx of adults had a related pair (adult vs child from the same family), higher proportion of Veillonella (OTU 2) and each unrelated pair (adult vs child not from the Lachnospiraceae (OTU 13) among others. Within same family) and between all adult or all child the nasopharynx, there were 82 statistically signifi- samples (Figure 3). In the nasopharynx, the cant OTUs, many more that were associated with microbial profiles of adults had the smallest adult samples than with child samples (Figure 4b). distance from one another. The profiles of child Of the three OTUs significantly associated with NP communities were most different from that of communities in young children, only Moraxella other children, the mean of which was signifi- (OTU 1) occurred in high abundance. OTUs sig- cantly different from the mean distances between nificantly associated with adult samples included adults and adults with unrelated children Staphylococcus (OTU 7), Lachnospiraceae (OTU (Po0.05). In the oropharynx, however, the micro- 13), Streptococcus (OTU 2), Anaerococcus (OTU bial profiles of children had the smallest distance 41) and Pseudomonas (OTU 27), among others (a from one another. Distances between adults and complete list is shown in Supplementary Table S5). adults and unrelated children were higher and Individual abundance histogram plots for all OTU significantly different from the child–child dis- shown in Figure 4a (NP) and 4b (OP) are presented tances (Po0.05). in Supplementary Figure S4 and S5, respectively, and complete lists of linear discriminant analysis effect sizes and P-values for OP and NP are presented in Supplementary Table S5 and S6, 0.5 * * respectively. * 0.4 * 0.3 Streptococcus diversity in the oropharynx and the

0.2 nasopharynx Some of the most prevalent bacteria found in this 0.1 study were the Streptococcus, especially in the

Weighted UniFrac Distance UniFrac Weighted oropharynx. Although OTU classification may have 0.0 underrepresented species as defined by classical taxonomic methods, all of the dominant populations except the Streptococcus were represented by a NP Adults OP Adults single OTU. It is difficult to resolve the taxonomy of NP Children OP Children the streptococci based on differences within the 16S NP Related AvC OP Related AvC NP Unrelated AvC OP Unrelated AvC rRNA gene and particularly based on only the short Figure 3 Distances between pairs of samples show community V3 region. We nonetheless endeavored to identify, differences between sites and ages of subjects. Weighted UniFrac which species of Streptococcus were represented by distances between each pair of samples were averaged for: related the most abundant OTUs using the pplacer phylo- adults and children, adult and child samples from all other genetic placement method. The Streptococcus phy- subjects (i.e., called ‘unrelated’), between all adults and between all children. Asterisks indicate significantly different means logeny created from full-length 16S rRNA gene tested with a Kruskal–Wallis test and a Dunn’s multiple sequences for species within the Human Oral comparison test, Po0.05). Microbiome Database (Supplementary Figures S6)

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1253 Adult Child <19 mo 1

0.1

0.01

0.001

0.0001

Relative Abundance 0.00001

0.000001 1 1 5 7 Rothia_10 raceae_34 Moraxella_1 Gemella_14 spi Raoultella_5 Streptococcus_2 Haemophilus_15 Anaerococcus_4 Pseudomonas_27 Staphylococcus_7 Bifidobacterium_7 7 Staphylococcus_6 Actinomycetales_62 Lachno Lachnospiraceae_13 Lachnospiraceae_32 Corynebacterium_5 6 Propionibacterium_66 Sphingobacterium_4

1

0.1

0.01

0.001

0.0001

Relative Abundance 0.00001

0.000001 9 6 6 richia_65 aceae_13 ot Neisseria_6 Gemella_14 pir Prevotella_3 Prevotella_2 Prevotella_37 Prevotella_71 Prevotella_23 Veillonella_59 Veillonella_12 Atopobium_63 Lept Actinomyces_29 Streptococcus_3 Haemophilus_15 Heamophilus_76 Granulicatella_25 Granulicatella_24 chnos Veillonellaceae_6 4 Campylobacter_4 6 Porphyromonas_3 0 La Fusobacteriaceae_3 Figure 4 Bacterial OTUs are significantly different between profiles from adults or young children at each site. (a) Bacterial OTUs from the nasopharynx and (b) the oropharynx. Only those OTUs with a total relative abundance across all samples of 40.11% are shown here along with an abundance histogram for each OTU (Supplementary Figures S4 and S5). A complete list of OTUs tested along with effect sizes and fdr adjusted P-values for each site is given in Supplementary Tables S5 and S6. was able to separate most species with the exception The taxa summaries from Figure 2a were reco- of a few closely related species such as Streptococ- loured to highlight the distribution of the three most cus oralis, Streptococcus mitis and S. pneumoniae; abundant Streptococcus OTUs, which were differ- and between Streptococcus intermedius, Strepto- ent in each group of samples (Figure 5). The coccus constellatus and Streptococcus anginosis oropharynx profiles of children had more mitis (known as the Streptococcus milleri/anginosus group than salivarius group Streptococcus, however, group). Despite having good separation between S. adults harboured more salivarius group than mitis salivarius and S. vestibularis in the reference group Streptococcus, as well as a small proportion phylogeny pplacer was unable to place many of of the milleri/anginosus group Streptococcus that sequences from isolated strains squarely with either was lacking in children. Likewise in the nasophar- one, therefore although OTU 2 was placed closer to ynx, child profiles carried more mitis group than S. vestibularis, we don’t have confidence about its salivarius group Streptococcus and lacked the mill- exact identity and believe that it likely represents a eri/anginosus group Streptococcus found in the mix of members of this clade (called salivarius adult nasopharynx. This group fluctuated greatly group here; Supplementary Figure S6). The second in the adult subset sometimes representing up to most highly abundant Streptococcus OTU (OTU 3) 23% of the community. This illustrates the diversity was placed within the clade containing S. mitis, S. of streptococci within the URT, which was masked pneumoniae and S. oralis, which we designated as by the taxa summary profiles at the genus level and the mitis group. Lastly, OTU 55 was placed with a suggests a role for this group in the maturation of the smaller more distinct clade made up of members of NP bacterial community, where either a lack of the Streptococcus milleri/anginosus group (S. angi- streptococci or dominance of one population is nosus, S. intermedius and S. constellatus) and likely replaced by a different distribution of the three represents one, or more, of these species. groups of species.

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1254 Quantitative culture adds context to molecular profiles associated bacteria have been cultured with these Direct sequencing of bacterial 16S rRNA genes methods from Sibley et al. (2011). Quantitative provided very detailed profiles for bacterial com- culturing of samples from 20 children and five munities in the URT of healthy people. Quantitative adults illustrated that total bacterial load varied culture, however, can provide absolute numbers of greatly with location and age (Figure 6a). Bacterial each cultivable bacterial population, providing load in the oropharynx was higher than that in the context to relative abundance profiles by adding nasopharynx for both adults and children, with total abundances. Although only a small proportion children carrying a higher total CFU ml À 1 than of environmental bacteria have been successfully adults. In the nasopharynx, cultures from children cultured (Stewart, 2012), B84% of human airway- showed a higher overall median number of

OTU 55 milleri/anginosus group

OTU 3 mitis group OTU 2 salivarius group

ChildAdult Adult Child Oropharyngeal Nasopharyngeal Figure 5 Diversity of Streptococcus OTUs differs between groups of samples. Taxa summaries from Figure 2a were recoloured for the three most abundant Streptococcus OTUs. The species group assigned to each OTU is derived from placement of each OTU within a phylogeny of Streptococcal species (Supplementary Figure S6) and from the taxonomy of cultured isolates, based on similarity with sequences within HOMD as shown by BLAST (Supplementary Table S8).

Oropharynx Nasopharynx 108 108 107 107 106 106 /ml 5 U 10 105 CF 4 CFU/ml 10 104 3 103 10 2 2 OP Child 10 10 OP Adult NP Child IIb NP Child IIa NP Child Ib NP Child Ia NP Adult

Unknown Proteobacteria Haemophilus influenzae Neisseria flava/mucosa Haemophilus parainfluenzae Neisseria flavescens Moraxella catarrhalis Pseudomonas sp.

Firmicutes Bacillus sp Streptococcus sp. Streptococcus parasanguinis

Relative Abundance Burkholderia sp pigrum Streptococcus salivarius group Gemella haemolysans Streptococcus mitis group Staphylococcus epidermidis Staphylococcus warneri Bacteriodetes Capnocytophaga Actinobacteria Actinomyces sp. Propionibacterium acnes C9607 C9148 C9167 C9238 C9148 C9149 C9262 C9506 C9660 C9662 C9670 C9707 C9167 C9238 C9289 C9699 C9560 C9607 C9666 C9714 C9692 C9705 C9668 C9162 A9238 A9167 A9148 A9662 A9692 A9607 A9148 A9238 A9167 Actinomyces odontolyticus Rothia sp. Corynebacterium durum Rothia mucilaginosa Corynebacterium sp./C. mucifaciens Child Ia Child Adult Child IIb Child IIa Child Ib Adult

Figure 6 Total bacterial biomass differs between samples. (a) Total bacterial counts for all bacterial species cultured. Samples from the oropharynx had the highest bacterial biomass, whereas samples from the nasopharynx of adults had the lowest bacterial biomass. None of the differences here were statistically significant based on a Kruskal–Wallis test and a Dunn’s multiple comparison’s test. (b) Range of total bacterial counts obtained for both sample site and age represented by geometric mean and 10–90 percent indicated by the whiskers. (c) Stacked bar chart of the relative abundance of each bacterial species/group cultured. Owing to the length of the sequence obtained from cultured isolates, species designations could be made. Members of the same Phylum are coloured with the same hue (Actinobacteria: pink/purple; Bacteroidetes: grey; Firmicutes: yellow/green/brown; Proteobacteria: blues; unknown: black).

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1255 CFU ml À 1 (1.5 Â 106 vs 4.2 Â 103) than for adults, a invasive disease, such as empyema and sepsis. In difference of almost three orders of magnitude. addition to the prevalence of pathogens, a compre- This is especially interesting in the context of data hensive investigation of microbes within the URT presented in Figure 2, as profiles that were grouped can provide insight into community structure, together into subclades had a different amount of aspects of which are important to disease suscept- total bacteria than those in other subclades. ibility. Children, who are at an increased risk of Figure 6b illustrates this point by ordering profiles respiratory infection, are known to have high according to their bacterial load, with those from carriage rates of pathogenic bacteria (Pelton, 2012), child NP samples split into their respective sub- however, their URT microbiome has not been fully clades. Next to orophayngeal samples, bacterial load characterized. Here, molecular and culture-based was highest in samples from subclade IIb (those methods were used to compare microbial commu- dominated by Moraxella), with between 106 and 107 nities from the nasopharynx and oropharynx of CFU ml À 1 Adult NP communities had the lowest, young children (aged 1–4.5 years, median 1.1) and with between 5 Â 102 and 104 CFU ml À 1 and samples their parents in hopes of describing the immature in subclade IIa (dominated by Corynebacterium and URT microbiota. Carnobacteriaceae) had an intermediate amount. As expected from molecular studies of the Interestingly, child samples within subclade Ib, nasopharynx and oropharynx (Keijser et al., 2008; which had profiles that resembled to those in adults, Lazerevic et al., 2009; Nasidze et al., 2009; Lemon had lower biomass than those of the other children et al., 2010; Charlson et al., 2011), within this study, in the study but a higher amount than adults. As these two niches haboured distinct bacterial com- there was a large disparity in the number of samples munities. OP bacterial communities of children and included in each group, a Kruskal–Wallis test found adults were dominated by streptococci (Figure 2) no statistically significant difference in the and the fact that bacterial profiles were so similar, CFU ml À 1 between groups. regardless of age, suggests that the colonization of Partial 16S rRNA gene sequences (B500–700 bp) the oropharynx occurs early and that communities from cultured isolates were also useful in suggesting are quickly dominated by Streptococcus. The oro- species level taxonomy for prevalent groups from pharynx is known to contain mainly Prevotella, direct sequenced molecular profiles in Figure 2. For Veillonella and Streptococcus (Keijser et al., 2008; instance, the only Moraxella species cultured was M. Lazerevic et al., 2009; Nasidze et al., 2009), with the catarrhalis and it was present in high abundance. latter being more prevalent here than previously Likewise, the only species of Carnobacteriaceae shown. Earlier studies may have underestimated the foundinNPsampleswasDolosigranulum pigrum, prevalence of this group, and Gram-positive bacteria which again was found in high numbers, providing a in general, as our DNA isolation protocol has been species designation for a group, which was only optimized to capture nearly all of the bacterial classified to the family level in the molecular profiles. diversity within the respiratory tract (Sibley et al., The Streptococcus species, however, were numerous 2011). A test of the differences in UniFrac distances and varied, and it seems likely that, for this group between samples showed that the bacterial commu- especially, OTU designations were not adequate to nities in the oropharynx of children were all most describe a species. For example, OTU 2, which was similar to one another and that adult profiles were identified above as belonging to the salivarius group, most different from one another or those of unre- likely contained more than one species, as these were lated children (Figure 3). Although few obvious isolated in abundance from the oropharynx and their taxonomic differences could be seen between adult partial 16S rRNA gene sequences were placed and child profiles from the oropharynx, several between S. vestibularis and S. salivarius along with OTUs were significantly different (Figure 4a). the representative sequence for OTU 2 on the Of interest is the fact that OTU 3, which was reference phylogeny (Supplementary Figure S6). characterized as a mitis group Streptococcus and Likewise, OTU 3 likely contained members from was highly abundant in all OP samples, was several species, including S. mitis, S. pneumoniae significantly associated with young children. Also, and S. oralis for the same reason (for a complete list of several OTUs belonging to the genera Prevotella had species, see Supplementary Table S7). significantly different proportions in the adult and the young child oropharynx. This is a complex genus with many species that may colonize people Discussion differently as they age. NP bacterial communities were heterogeneous but Characterization of the microbial communities in overall adult samples had higher diversity the URT is important as, even in healthy people, it (Figure 1a) and lower bacterial abundance can harbour both pathogens and potentially patho- (Figure 6b). Samples from children contained a genic commensal organisms (Garcia-Rodriguez and range of bacteria, with some dominated by Proteo- Fresnadillo Martinez, 2002; Bakaletz, 2004; Liu bacteria (Moraxella), others containing a large et al., 2013). These are responsible for acute proportion of Actinobacteria and Firmicutes (Cor- respiratory infections such as and ynebacterium and D. pigrum), and others resembling

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1256 adult NP swabs. Interestingly, within the Moraxella- richness and diversity estimates (Haegeman et al., dominated group, the evidence of competitive 2013). Why child NP communities should have such exclusion between Haemophilus and Streptococcus high bacterial loads is unknown but could be related was seen in a small number of samples, an to increased immune tolerance (Jaspan et al., 2006) observation previously made during studies on the in the immature nasopharynx. Longitudinal quanti- interactions of individual URT resident pathogens tative culture studies designed to capture commen- (Lijek and Weiser, 2012). A small number of samples sal bacterial diversity are needed to answer this were dominated by streptococci and hence more completely. resembled swabs from the oropharynx. When all Maturation of the host-associated microbial com- samples were clustered based on phylogenetic munity is known to occur in gut communities where differences, four groups emerged two of which were progression from colonization to climax and stabi- enriched with samples from the nasopharynx of lity (Adlerberth and Wold, 2009), as well as children (Figure 2c). Similarly, an ordination of NP refraction of this community to disturbance microbial profiles from children in the Netherlands (Dethlefsen et al., 2008), has been studied. It is also, of similar age revealed four clusters, three domi- then, possible that a similar effect occurs in the nated by a single OTU (Moraxella, Haemophilus or airways, where a newborn child is colonized by Streptococcus) and one, which was mixed (Bogaert bacteria from their surroundings and that these et al., 2011). It is unknown if the fourth mixed group bacterial populations succeed one another until an of child samples resembled adult profiles, as none equilibrium is reached and a stable community were included, however, this aligns well with our persists. In the gut, a healthy stable microbiota data, suggesting that NP bacterial communities of protects a person from disease (Round and young children often have low diversity owing to Mazmanian, 2009), likewise a healthy, stable URT the dominance by a small number of bacterial microbiota may have a similar role. In fact, the species. The prevalence of Corynebacterium and D. microbial composition has been shown to regulate pigrum instead of Haemophilus could be due to immune modulation in response to viral pathogens seasonal or geographical influence on bacterial in the URT (Ichinohe et al., 2011). What impact a carriage, yet may also be due to better representation person’s immune system has on the composition of of Gram-positive bacteria in our work. The NP the NP microbial community is not yet known. The profiles of adults were most similar to one another profiles presented here of children whose nasophar- and distances were greatest between communities in ynx had high bacterial load of one or two dominant children and between those in children and unre- organisms (that is, M. catarrhalis, D. pigrum and lated adults (Figure 3). The distances between Corynebacterium sp.) could be an illustration of the children does not take into account the fact that immature NP bacterial community that is still there were two different groups of children. dynamic and will eventually settle to one with An important finding from this study was that lower biomass that is more stable and resembles not only did child NP samples differ in bacterial what is found in mid-aged adults. Whether the membership, but that they also had different total children harbouring high Moraxella numbers will bacterial numbers (Figure 6). Overall, bacterial have different health outcomes from the others is numbers differed by several orders of magnitude impossible to say without longitudinal studies, between adult and child samples, with children however, this has been true for rates of acute otitis always carrying a higher number than adults at each media in young children (Faden et al., 1994). site. Within each group of child NP samples, Children, along with the elderly, are at the highest differences in the bacterial load suggests further risk for respiratory infection. In a similar study from differences between the microbial communities our laboratory, the nasal microbiota within elderly within these individuals that were not apparent nursing home patients were seen to have very little with molecular profiles alone. This observation does similarity between people, suggesting a loss of echo alpha diversity, calculated from molecular community order when compared with mid-aged profiles (Figure 1a), where NP samples from young adults (Whelan et al., 2014). Differences in bacterial children had lower species richness and community community structure within the URT of children diversity, compared with adult profiles. These and the elderly, when compared with mid-aged observations make sense in the context of commu- adults, may be contributing to the susceptibility of nity maturation, as lower biomass and higher people within these two age groups to infection. diversity are hallmarks of a climax community The Streptococcus have a range of phenotypes and (Loucks, 1970), which is stable (McCann, 2000). behaviours in association with human mucosal sites. Caution needs to be taken with molecular profiles Whereas some, like S. pneumoniae can behave based on DNA and high throughput sequencing as, pathogenically in some people (Walker et al., 2013), when there are differences in bacterial biomass, an others such as S. salivarius and S. mitis are found equal depth of sampling will always uncover both commonly in healthy human mouths. Not surprisingly, real and spurious low abundance OTUs (Schloss this group was prevalent here, found abundantly in OP et al., 2011) and microbial community richness samples of both children and adults, seen to dominate becomes difficult to estimate without inflating a small number of immature NP samples and seen

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1257 often in mature NP samples. At the OTU level, the References nasopharynx and oropharynx contained two abundant and one less abundant OTU with several other low Adlerberth I, Wold AE. (2009). Establishment of the abundant streptococci. However, due to the similarity gut microbiota in western infants. Acta paediatr 98: 229–238. of the 16S rRNA gene within these short sequences, it Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ. is difficult to resolve species identification. In fact (1990). Basic local alignment search tool. J Mol Biol when longer sequences, obtained from cultured iso- 215: 403–410. lates, were classified, hidden diversity within Strepto- Bakaletz LO. (2004). Developing animal models for coccal OTUs was revealed: OTUs 2 and 3 and 55 likely polymicrobial diseases. Nat Rev Microbiol 2: 552–568. contained a mix of salivarius group, mitis group and Bartram AK, Lynch MDJ, Stearns JC, Moreno-Hagelsieb G, milleri/anginosus group streptococci, respectively Neufeld JD. (2011). Generation of multimillion- (Supplementary Figure S6). Therefore, from culture, sequence 16S rRNA gene libraries from complex we can identify more than one dominant species microbial communities by assembling paired-end within each Streptococcus OTU,afactthatwasnot Illumina reads. Appl Environ Microbiol 77: 3846–3852. seen with the other abundant genera found. This was Benjamini Y, Hochberg Y. (1995). Controlling the false not surprising given both the diversity of streptococci discovery rate: a practical and powerful approach to traditionally found in the URT and the difficulty in multiple testing. J Royal Stat Soc Ser B 57: 289–300. distinguishing between streptococcal species with Blaser MJ, Falkow S. (2009). What are the consequences DNA profiling. Even culture methods based on colony of the disappearing human microbiota? Nat Rev morphology have been shown to underestimate the Microbiol 7: 887–894. diversity of closely related species (Sibley et al., 2011), Bogaert D, Van Belkum A, Sluijter M, Luijendijk A, and it is likely that the Streptococcus diversity is De Groot R, Ru¨ mke HC et al. (2004). Colonisation greater than estimated for these samples. by Streptococcus pneumoniae and Staphylococcus Several studies of URT colonization by bacteria, aureus in healthy children. Lancet 363: 1871–1872. Bogaert D, Keijser B, Huse S, Rossen J, Veenhoven R, while measuring carriage rates of specific pathogens van Gils E et al. (2011). Variability and diversity of in the juvenile population, have described popula- nasopharyngeal microbiota in children: a metage- tions using mainly clinical culture methods nomic analysis. PLoS One 6: e17035. (Vaneechoutte et al., 1990; Bogaert et al., 2004; Caporaso GJ, Kuczynski J, Stombaugh J, Bittinger K, Regev-Yochay et al., 2004). Although these studies Bushman FD, Costello EK et al. (2010). QIIME allows describe the presence of specific pathogens, they analysis of high-throughput community sequencing neglect to describe the remainder of the bacterial data. Nat Methods 7: 335–336. community, which may contribute to polymicrobial Charlson ES, Bittinger K, Haas AR, Fitzgerald AS, Frank I, lower airway infection (Bakaletz, 2004; Han et al., Yadav A et al. (2011). Topographical continuity 2012; Huang et al., 2012; Dickson et al., 2013; Huang of bacterial populations in the healthy human respiratory tract. Am J of Respir Crit Care Med 184: et al., 2014). Many of the bacterial strains living 957–963. benignly in the healthy URT can act pathogenically Chen T, Yu WH, Izard J, Baranova O, Lakshmanan A, in the lower airways and elsewhere. Here, the Dewhirst F. (2010). The human oral microbiome microbial landscape of the URT in healthy children database: a web accessible resource for investigating and their parents presents aspects such as commu- oral microbe taxonomic and genomic information. nity membership and bacterial load that are likely Database 2010: baq013. important for child health and disease, as well as DeSantis TZ, Hugenholtz P, Larsen N, Rojas M, Brodie EL, highlighting further questions about when and how Keller K et al. (2006). Greengenes, a chimera-checked the microbiota matures in the nasopharynx and 16S rRNA gene database and workbench compatible what role the host immune system has in this with ARB. Appl Environ Microbiol 72: 5069–5072. Dethlefsen L, Huse S, Sogin ML, Relman DA. (2008). process. The pervasive effects of an antibiotic on the human gut microbiota, as revealed by deep 16S rRNA sequencing. PLoS Biol 6: e280. Conflict of Interest Dickson RP, Erb-Downward JR, Huffnagle GB. (2013). The The authors declare no conflict of interest. role of the bacterial microbiome in lung disease. Expert Rev Respir Med 7: 245–257. Dunne EM, Smith-Vaughan HC, Robins-Browne RM, Acknowledgements Mulholland EK, Satzke C. (2013). Nasopharyngeal microbial interactions in the era of pneumococcal We would like to thank Laura Rossi and Christine King for conjugate vaccination. Vaccine 31: 2333–2342. their contribution to the Illumina library preps and Edgar RC. (2004). MUSCLE: a multiple sequence sequencing and their contributions to this manuscript. alignment with high accuracy and high throughput. This work was funded by the Canadian Institute for Health BMC Bioinformatics 5: 113. Research. MGS was supported as a Canada Research Chair Faden H, Harabuchi Y, Hong JJTonawanda/Williamsville in Interdisciplinary Microbiome Research. DMEB was Pediatrics (1994). Epidemiology of Moraxella supported in part by the McMaster Immunology Research catarrhalis in children during the first 2 years of Centre. DMEB and MGS were supported in part by the MG life: relationship to otitis media. J Infect Dis 169: DeGroote Institute for Infectious Disease Research. 1312–1317.

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1258 Faust K, Sathirapongsasuti JF, Izard J, Segata N, Gevers D, Lijek RS, Weiser JN. (2012). Co-infection subverts mucosal Raes J et al. (2012). Microbial co-occurrence immunity in the upper respiratory tract. Curr Opin relationships in the human microbiome. PLoS Comp Immunol 24: 417–423. Biol 8: e1002606. Liu H, Wang H, Huang M, Mei Y, Gu B, Wu R et al. (2013). Feldman C, Anderson R. (2013). Cigarette smoking and Analysis of antimicrobial resistance and class 1 mechanisms of susceptibility to infections of the integrons among strains from upper respiratory tract respiratory tract and other organ systems. J Infect 67: of healthy adults. J Thor Dis 5: 149. 169–184. Liu L, Johnson HL, Cousens S, Perin J, Scott S, Lawn JE et al. Garcia-Rodriguez JA, Fresnadillo Martinez MJ. (2002). (2012). Global, regional, and national causes of Dynamics of nasopharyngeal colonization by potential child mortality: an updated systematic analysis respiratory pathogens. J Antimicrob Chemother 50: for 2010 with time trends since 2000. Lancet 379: 59–74. 2151–2161. Haegeman B, Hamelin J, Moriarty J, Neal P, Dushoff J, Liu WT, Marsh TL, Cheng H, Forney LJ. (1997). Weitz JS. (2013). Robust estimation of microbial Characterization of microbial diversity by determining diversity in theory and in practice. ISME J 7: terminal restriction fragment length polymorphisms of 1092–1101. genes encoding 16S rRNA. Appl Environ Microbiol 63: Han MK, Huang YJ, LiPuma JJ, Boushey HA, Boucher RC, 4516–4522. Cookson WO et al. (2012). Significance of the Loucks OL. (1970). Evolution of diversity, efficiency, microbiome in obstructive lung disease. Thorax 67: and community stability. Am Zool 10: 17–25. 456–463. Lynch SV. (2013). Probiotic bacteria and their effect on Huang YJ, Lynch SV, Wiener-Kronish JP. (2012). From human health and well-being. In: Guarino A, Quigley microbe to microbiota: considering microbial commu- E, & Walker W (eds). World Rev Nutr Diet, Vol 107. nity composition in infections and airway diseases. Karger: Basel, Switzerland, pp 64–71. Am J Resp Crit Care Med 185: 691–692. Margolis E, Yates A, Levin BR. (2010). The ecology of nasal Huang YJ, Sethi S, Murphy T, Nariya S, Boushey HA, colonization of Streptococcus pneumoniae, Haemo- Lynch SV. (2014). Airway microbiome dynamics philus influenzae and Staphylococcus aureus: the role in exacerbations of chronic obstructive pulmonary of competition and interactions with host’s immune disease. J Clin Microbiol 52: 2813–2823. response. BMC Microbiol 10: 59. Ichinohe T, Pang IK, Kumamoto Y, Peaper DR, Ho JH, Martin M. (2011). Cutadapt removes adapter sequences Murray TS et al. (2011). Microbiota regulates immune from high-throughput sequencing reads. EMBnet J defense against respiratory tract influenza A virus 17: 10. infection. PNAS 108: 5354–5359. Masella AP, Bartram A, Truszkowski JM, Brown DG, Jaspan HB, Lawn SD, Safrit JT, Bekker LG. (2006). Neufeld JD. (2012). Pandaseq: paired-end The maturing immune system: Implications for assembler for Illumina sequences. BMC Bioinformatics development and testing HIV-1 vaccines for children 13: 31. and adolescents. Aids 20: 483–494. Matsen FA, Kodner RB, Armbrust EV. (2010). pplacer: Keijser B, Zaura E, Huse S, van der Vossen J, Schuren F, linear time maximum-likelihood and Bayesian Montijn R et al. (2008). Pyrosequencing analysis phylogenetic placement of sequences onto a fixed of the oral microflora of healthy adults. J Dent Res reference tree. BMC bioinformatics 11: 538. 87: 1016–1020. McCann KS. (2000). The diversity-stability debate. Nature Kellner JD, Scheifele D, Vanderkooi OG, MacDonald J, 405: 228–233. Church DL, Tyrrell GJ. (2008). Effects of routine McMurdie PJ, Holmes S. (2013). Phyloseq: an R package infant vaccination with the 7-valent pneumococcal for reproducible interactive analysis and graphics of conjugate vaccine on nasopharyngeal colonization microbiome census data. PloS One 8: e61217. with Streptococcus pneumoniae in children Mizgerd J. (2006). Lung infection– a public health priority. in Calgary, Canada. Pediatr Infect Dis J 27: PLoS Med 3: e76. 526–532. Mizgerd J. (2008). Acute lower respiratory tract infection. Kelly D, Conway S, Aminov R. (2005). Commensal gut N Engl J Med 358: 716–727. bacteria: mechanisms of immune modulation. Trends Nasidze I, Li J, Quinque D, Tang K, Stoneking M. (2009). Immunol 26: 326–333. Global diversity in the human salivary microbiome. Lathrop SK, Bloom SM, Rao SM, Nutsch K, Lio CW, Genome Res 19: 636–643. Santacruz N et al. (2011). Peripheral education of the O’Brien K, Nohynek H. (2003). Report from a WHO immune system by colonic commensal microbiota. working group: standard method for detecting upper Nature 478: 250–254. respiratory carriage of Streptococcus pneumoniae. Laupland KB, Davies HD, Low DE, Schwartz B, Green K, Pediatr Infect Dis J 22: e1–e11. McGeer A. (2000). Invasive group A Streptococcal Pelton SI. (2012). Regulation of bacterial trafficking in the disease in children and association with nasopharynx. Paediatr Resp Rev 13: 150–153. varicella-zoster virus Infection. Pediatrics 105: Rambaut A, Drummond ATracer v1.6 (2009). Available e60–e60. athttp://beast.bio.ed.ac.uk/software/Tracer. Lazarevic V, Whiteson K, Huse S, Hernandez D, Farinelli L, Regev-Yochay G, Dagan R, Raz M, Carmeli Y, Shainberg B, Osteras M et al. (2009). Metagenomic study of the oral Derazne E et al. (2004). Association between carriage microbiota by Illumina high-throughput sequencing. of Streptococcus pneumoniae and Staphylococcus J Microbiol Methods 79: 266–271. aureus in children. JAMA 292: 716–720. Lemon KP, Klepac-ceraj V, Schiffer HK, Brodie EL, Lynch SV, Ricketson LJ, Wood ML, Vanderkooi OG, MacDonald JC, Kolter R. (2010). Comparative analyses of the bacterial Kellner JD. (2014). Trends in asymptomatic nasopah- microbiota of the human nostril and oropharynx. mBio ryngeal colonization with Streptococcus pneumoniae 1: e00129–10. after introduction of the 13- valent pneumococcal

The ISME Journal Changes in the microbiota of the upper airways with age JC Stearns et al 1259 conjugate vaccine in Calgary, Canada. Pediatr Infect Stamatakis A. (2006). RAxML-VI-HPC: maximum Dis J 33: 724–730. likelihood-based phylogenetic analyses with Ronquist F, Huelsenbeck JP. (2003). MrBayes 3: Bayesian thousands of taxa and mixed models. Bioinformatics phylogenetic inference under mixed models. 22: 2688–2690. Bioinformatics 19: 1572–1574. Starkey MR, Nguyen DH, Kim RY, Nair PM, Brown AC, Round JL, Mazmanian SK. (2009). The gut microbiota Essifie AT et al. (2013). Programming of the lung in shapes intestinal immune responses during health early life by bacterial infections predisposes to chronic and disease. Nat Rev Immunol 9: 313–323. respiratory disease. Clin Obstet Gynecol 56: 566–576. Schloss PD, Gevers D, Westcott SL. (2011). Reducing the Stewart EJ. (2012). Growing unculturable bacteria. effects of PCR amplification and sequencing artifacts J Bacteriol 194: 4151–4160. on 16S rRNA-based studies. PLoS One 6: e27310. Vaneechoutte M, Verschraegen G, Claeys G, Weise B, Segata N, Izard J, Walron L, Gevers D, Miropolsky L, Van den Abeele AM. (1990). Respiratory tract carrier Garrett W et al. (2011). Metagenomic biomarker rates of Moraxella (Branhamella) catarrhalis in adults discovery and explanation. Gen Biol 12: R60. and children and interpretation of the isolation Shannon CE, Weaver W. (1949). The Mathematical Theory of of M. catarrhalis from sputum. J Clin Microbiol 28: Communication. University of Illinois Press: Urbana 5. 2674–2680. Sibley CD, Church DL, Surette MG, Dowd SE, Parkins MD. Walker CLF, Rudan I, Liu L, Nair H, Theodoratou E, Bhutta Z (2012). Pyrosequencing reveals the complex et al. (2013). Global burden of childhood pneumonia polymicrobial nature of invasive pyogenic infections: and diarrhoea. Lancet 381: 1405–1416. microbial constituents of empyema, liver abscess, and Wang Q, George M, Tiedje J, Cole J. (2007). Naive Bayesian intracerebral abscess. Eur J Clin Microbiol Infect Dis classifier for rapid assignment of rRNA sequences into 31: 2679–2691. the new bacterial taxonomy. App Environ Microbiol Sibley CD, Grinwis ME, Field TR, Eshaghurshan CS, 73: 5261–5267. Faria MM, Dowd SE et al. (2011). Culture enriched Whelan F, Verschoor C, Stearns J, Rossi L, Johnstone J, molecular profiling of the cystic fibrosis airway Surette M et al. (2014). The loss of topography in the microbiome. PLoS One 6: 11. microbial communities of the upper respiratory tract Sibley CD, Parkins MD, Rabin HR, Duan K, Norgaard JC, in the elderly. Ann Am Thor Soc 11: 513–521. Surette MG. (2008). A polymicrobial perspective of Ye Y. (2011). Identification and Quantification of pulmonary infections exposes an enigmatic pathogen Abundant Species from Pyrosequences of 16S rRNA in cystic fibrosis patients. PNAS 105: 15070–15075. by Consensus Alignment. Proceedings (IEEE Int Conf Silvestro D, Michalak I. (2012). raxmlGUI: a graphical front- Bioinformatics Biomed). 153–157; doi: 10.1109/BIBM. end for RAxML. Org Divers and Evol 12: 335–337. 2010.5706555.

Supplementary Information accompanies this paper on The ISME Journal website (http://www.nature.com/ismej)

The ISME Journal